fs/dlm/dlm_internal.h

   1 /******************************************************************************
   2 *******************************************************************************
   3 **
   4 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   5 **  Copyright (C) 2004-2005 Red Hat, Inc.  All rights reserved.
   6 **
   7 **  This copyrighted material is made available to anyone wishing to use,
   8 **  modify, copy, or redistribute it subject to the terms and conditions
   9 **  of the GNU General Public License v.2.
  10 **
  11 *******************************************************************************
  12 ******************************************************************************/
  13
  14 #ifndef __DLM_INTERNAL_DOT_H__
  15 #define __DLM_INTERNAL_DOT_H__
  16
  17 /*
  18  * This is the main header file to be included in each DLM source file.
  19  */
  20
  21 #include <linux/module.h>
  22 #include <linux/slab.h>
  23 #include <linux/sched.h>
  24 #include <linux/types.h>
  25 #include <linux/ctype.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/vmalloc.h>
  28 #include <linux/list.h>
  29 #include <linux/errno.h>
  30 #include <linux/random.h>
  31 #include <linux/delay.h>
  32 #include <linux/socket.h>
  33 #include <linux/kthread.h>
  34 #include <linux/kobject.h>
  35 #include <linux/kref.h>
  36 #include <linux/kernel.h>
  37 #include <linux/jhash.h>
  38 #include <linux/miscdevice.h>
  39 #include <linux/mutex.h>
  40 #include <asm/semaphore.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/dlm.h>
  44
  45 #define DLM_LOCKSPACE_LEN       64
  46
  47 /* Size of the temp buffer midcomms allocates on the stack.
  48    We try to make this large enough so most messages fit.
  49    FIXME: should sctp make this unnecessary? */
  50
  51 #define DLM_INBUF_LEN           148
  52
  53 struct dlm_ls;
  54 struct dlm_lkb;
  55 struct dlm_rsb;
  56 struct dlm_member;
  57 struct dlm_lkbtable;
  58 struct dlm_rsbtable;
  59 struct dlm_dirtable;
  60 struct dlm_direntry;
  61 struct dlm_recover;
  62 struct dlm_header;
  63 struct dlm_message;
  64 struct dlm_rcom;
  65 struct dlm_mhandle;
  66
  67 #define log_print(fmt, args...) \
  68         printk(KERN_ERR "dlm: "fmt"\n" , ##args)
  69 #define log_error(ls, fmt, args...) \
  70         printk(KERN_ERR "dlm: %s: " fmt "\n", (ls)->ls_name , ##args)
  71
  72 #define DLM_LOG_DEBUG
  73 #ifdef DLM_LOG_DEBUG
  74 #define log_debug(ls, fmt, args...) log_error(ls, fmt, ##args)
  75 #else
  76 #define log_debug(ls, fmt, args...)
  77 #endif
  78
  79 #define DLM_ASSERT(x, do) \
  80 { \
  81   if (!(x)) \
  82   { \
  83     printk(KERN_ERR "\nDLM:  Assertion failed on line %d of file %s\n" \
  84                "DLM:  assertion:  \"%s\"\n" \
  85                "DLM:  time = %lu\n", \
  86                __LINE__, __FILE__, #x, jiffies); \
  87     {do} \
  88     printk("\n"); \
  89     BUG(); \
  90     panic("DLM:  Record message above and reboot.\n"); \
  91   } \
  92 }
  93
  94 #define DLM_FAKE_USER_AST ERR_PTR(-EINVAL)
  95
  96
  97 struct dlm_direntry {
  98         struct list_head        list;
  99         uint32_t                master_nodeid;
 100         uint16_t                length;
 101         char                    name[1];
 102 };
 103
 104 struct dlm_dirtable {
 105         struct list_head        list;
 106         rwlock_t                lock;
 107 };
 108
 109 struct dlm_rsbtable {
 110         struct list_head        list;
 111         struct list_head        toss;
 112         rwlock_t                lock;
 113 };
 114
 115 struct dlm_lkbtable {
 116         struct list_head        list;
 117         rwlock_t                lock;
 118         uint16_t                counter;
 119 };
 120
 121 /*
 122  * Lockspace member (per node in a ls)
 123  */
 124
 125 struct dlm_member {
 126         struct list_head        list;
 127         int                     nodeid;
 128         int                     weight;
 129 };
 130
 131 /*
 132  * Save and manage recovery state for a lockspace.
 133  */
 134
 135 struct dlm_recover {
 136         struct list_head        list;
 137         int                     *nodeids;
 138         int                     node_count;
 139         uint64_t                seq;
 140 };
 141
 142 /*
 143  * Pass input args to second stage locking function.
 144  */
 145
 146 struct dlm_args {
 147         uint32_t                flags;
 148         void                    *astaddr;
 149         long                    astparam;
 150         void                    *bastaddr;
 151         int                     mode;
 152         struct dlm_lksb         *lksb;
 153 };
 154
 155
 156 /*
 157  * Lock block
 158  *
 159  * A lock can be one of three types:
 160  *
 161  * local copy      lock is mastered locally
 162  *                 (lkb_nodeid is zero and DLM_LKF_MSTCPY is not set)
 163  * process copy    lock is mastered on a remote node
 164  *                 (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is not set)
 165  * master copy     master node's copy of a lock owned by remote node
 166  *                 (lkb_nodeid is non-zero and DLM_LKF_MSTCPY is set)
 167  *
 168  * lkb_exflags: a copy of the most recent flags arg provided to dlm_lock or
 169  * dlm_unlock.  The dlm does not modify these or use any private flags in
 170  * this field; it only contains DLM_LKF_ flags from dlm.h.  These flags
 171  * are sent as-is to the remote master when the lock is remote.
 172  *
 173  * lkb_flags: internal dlm flags (DLM_IFL_ prefix) from dlm_internal.h.
 174  * Some internal flags are shared between the master and process nodes;
 175  * these shared flags are kept in the lower two bytes.  One of these
 176  * flags set on the master copy will be propagated to the process copy
 177  * and v.v.  Other internal flags are private to the master or process
 178  * node (e.g. DLM_IFL_MSTCPY).  These are kept in the high two bytes.
 179  *
 180  * lkb_sbflags: status block flags.  These flags are copied directly into
 181  * the caller's lksb.sb_flags prior to the dlm_lock/dlm_unlock completion
 182  * ast.  All defined in dlm.h with DLM_SBF_ prefix.
 183  *
 184  * lkb_status: the lock status indicates which rsb queue the lock is
 185  * on, grant, convert, or wait.  DLM_LKSTS_ WAITING/GRANTED/CONVERT
 186  *
 187  * lkb_wait_type: the dlm message type (DLM_MSG_ prefix) for which a
 188  * reply is needed.  Only set when the lkb is on the lockspace waiters
 189  * list awaiting a reply from a remote node.
 190  *
 191  * lkb_nodeid: when the lkb is a local copy, nodeid is 0; when the lkb
 192  * is a master copy, nodeid specifies the remote lock holder, when the
 193  * lkb is a process copy, the nodeid specifies the lock master.
 194  */
 195
 196 /* lkb_ast_type */
 197
 198 #define AST_COMP                1
 199 #define AST_BAST                2
 200
 201 /* lkb_status */
 202
 203 #define DLM_LKSTS_WAITING       1
 204 #define DLM_LKSTS_GRANTED       2
 205 #define DLM_LKSTS_CONVERT       3
 206
 207 /* lkb_flags */
 208
 209 #define DLM_IFL_MSTCPY          0x00010000
 210 #define DLM_IFL_RESEND          0x00020000
 211 #define DLM_IFL_DEAD            0x00040000
 212 #define DLM_IFL_USER            0x00000001
 213 #define DLM_IFL_ORPHAN          0x00000002
 214
 215 struct dlm_lkb {
 216         struct dlm_rsb          *lkb_resource;  /* the rsb */
 217         struct kref             lkb_ref;
 218         int                     lkb_nodeid;     /* copied from rsb */
 219         int                     lkb_ownpid;     /* pid of lock owner */
 220         uint32_t                lkb_id;         /* our lock ID */
 221         uint32_t                lkb_remid;      /* lock ID on remote partner */
 222         uint32_t                lkb_exflags;    /* external flags from caller */
 223         uint32_t                lkb_sbflags;    /* lksb flags */
 224         uint32_t                lkb_flags;      /* internal flags */
 225         uint32_t                lkb_lvbseq;     /* lvb sequence number */
 226
 227         int8_t                  lkb_status;     /* granted, waiting, convert */
 228         int8_t                  lkb_rqmode;     /* requested lock mode */
 229         int8_t                  lkb_grmode;     /* granted lock mode */
 230         int8_t                  lkb_bastmode;   /* requested mode */
 231         int8_t                  lkb_highbast;   /* highest mode bast sent for */
 232
 233         int8_t                  lkb_wait_type;  /* type of reply waiting for */
 234         int8_t                  lkb_ast_type;   /* type of ast queued for */
 235
 236         struct list_head        lkb_idtbl_list; /* lockspace lkbtbl */
 237         struct list_head        lkb_statequeue; /* rsb g/c/w list */
 238         struct list_head        lkb_rsb_lookup; /* waiting for rsb lookup */
 239         struct list_head        lkb_wait_reply; /* waiting for remote reply */
 240         struct list_head        lkb_astqueue;   /* need ast to be sent */
 241         struct list_head        lkb_ownqueue;   /* list of locks for a process */
 242
 243         char                    *lkb_lvbptr;
 244         struct dlm_lksb         *lkb_lksb;      /* caller's status block */
 245         void                    *lkb_astaddr;   /* caller's ast function */
 246         void                    *lkb_bastaddr;  /* caller's bast function */
 247         long                    lkb_astparam;   /* caller's ast arg */
 248 };
 249
 250
 251 struct dlm_rsb {
 252         struct dlm_ls           *res_ls;        /* the lockspace */
 253         struct kref             res_ref;
 254         struct mutex            res_mutex;
 255         unsigned long           res_flags;
 256         int                     res_length;     /* length of rsb name */
 257         int                     res_nodeid;
 258         uint32_t                res_lvbseq;
 259         uint32_t                res_hash;
 260         uint32_t                res_bucket;     /* rsbtbl */
 261         unsigned long           res_toss_time;
 262         uint32_t                res_first_lkid;
 263         struct list_head        res_lookup;     /* lkbs waiting on first */
 264         struct list_head        res_hashchain;  /* rsbtbl */
 265         struct list_head        res_grantqueue;
 266         struct list_head        res_convertqueue;
 267         struct list_head        res_waitqueue;
 268
 269         struct list_head        res_root_list;      /* used for recovery */
 270         struct list_head        res_recover_list;   /* used for recovery */
 271         int                     res_recover_locks_count;
 272
 273         char                    *res_lvbptr;
 274         char                    res_name[1];
 275 };
 276
 277 /* find_rsb() flags */
 278
 279 #define R_MASTER                1       /* only return rsb if it's a master */
 280 #define R_CREATE                2       /* create/add rsb if not found */
 281
 282 /* rsb_flags */
 283
 284 enum rsb_flags {
 285         RSB_MASTER_UNCERTAIN,
 286         RSB_VALNOTVALID,
 287         RSB_VALNOTVALID_PREV,
 288         RSB_NEW_MASTER,
 289         RSB_NEW_MASTER2,
 290         RSB_RECOVER_CONVERT,
 291         RSB_LOCKS_PURGED,
 292 };
 293
 294 static inline void rsb_set_flag(struct dlm_rsb *r, enum rsb_flags flag)
 295 {
 296         __set_bit(flag, &r->res_flags);
 297 }
 298
 299 static inline void rsb_clear_flag(struct dlm_rsb *r, enum rsb_flags flag)
 300 {
 301         __clear_bit(flag, &r->res_flags);
 302 }
 303
 304 static inline int rsb_flag(struct dlm_rsb *r, enum rsb_flags flag)
 305 {
 306         return test_bit(flag, &r->res_flags);
 307 }
 308
 309
 310 /* dlm_header is first element of all structs sent between nodes */
 311
 312 #define DLM_HEADER_MAJOR        0x00020000
 313 #define DLM_HEADER_MINOR        0x00000001
 314
 315 #define DLM_MSG                 1
 316 #define DLM_RCOM                2
 317
 318 struct dlm_header {
 319         uint32_t                h_version;
 320         uint32_t                h_lockspace;
 321         uint32_t                h_nodeid;       /* nodeid of sender */
 322         uint16_t                h_length;
 323         uint8_t                 h_cmd;          /* DLM_MSG, DLM_RCOM */
 324         uint8_t                 h_pad;
 325 };
 326
 327
 328 #define DLM_MSG_REQUEST         1
 329 #define DLM_MSG_CONVERT         2
 330 #define DLM_MSG_UNLOCK          3
 331 #define DLM_MSG_CANCEL          4
 332 #define DLM_MSG_REQUEST_REPLY   5
 333 #define DLM_MSG_CONVERT_REPLY   6
 334 #define DLM_MSG_UNLOCK_REPLY    7
 335 #define DLM_MSG_CANCEL_REPLY    8
 336 #define DLM_MSG_GRANT           9
 337 #define DLM_MSG_BAST            10
 338 #define DLM_MSG_LOOKUP          11
 339 #define DLM_MSG_REMOVE          12
 340 #define DLM_MSG_LOOKUP_REPLY    13
 341
 342 struct dlm_message {
 343         struct dlm_header       m_header;
 344         uint32_t                m_type;         /* DLM_MSG_ */
 345         uint32_t                m_nodeid;
 346         uint32_t                m_pid;
 347         uint32_t                m_lkid;         /* lkid on sender */
 348         uint32_t                m_remid;        /* lkid on receiver */
 349         uint32_t                m_parent_lkid;
 350         uint32_t                m_parent_remid;
 351         uint32_t                m_exflags;
 352         uint32_t                m_sbflags;
 353         uint32_t                m_flags;
 354         uint32_t                m_lvbseq;
 355         uint32_t                m_hash;
 356         int                     m_status;
 357         int                     m_grmode;
 358         int                     m_rqmode;
 359         int                     m_bastmode;
 360         int                     m_asts;
 361         int                     m_result;       /* 0 or -EXXX */
 362         char                    m_extra[0];     /* name or lvb */
 363 };
 364
 365
 366 #define DLM_RS_NODES            0x00000001
 367 #define DLM_RS_NODES_ALL        0x00000002
 368 #define DLM_RS_DIR              0x00000004
 369 #define DLM_RS_DIR_ALL          0x00000008
 370 #define DLM_RS_LOCKS            0x00000010
 371 #define DLM_RS_LOCKS_ALL        0x00000020
 372 #define DLM_RS_DONE             0x00000040
 373 #define DLM_RS_DONE_ALL         0x00000080
 374
 375 #define DLM_RCOM_STATUS         1
 376 #define DLM_RCOM_NAMES          2
 377 #define DLM_RCOM_LOOKUP         3
 378 #define DLM_RCOM_LOCK           4
 379 #define DLM_RCOM_STATUS_REPLY   5
 380 #define DLM_RCOM_NAMES_REPLY    6
 381 #define DLM_RCOM_LOOKUP_REPLY   7
 382 #define DLM_RCOM_LOCK_REPLY     8
 383
 384 struct dlm_rcom {
 385         struct dlm_header       rc_header;
 386         uint32_t                rc_type;        /* DLM_RCOM_ */
 387         int                     rc_result;      /* multi-purpose */
 388         uint64_t                rc_id;          /* match reply with request */
 389         char                    rc_buf[0];
 390 };
 391
 392 struct rcom_config {
 393         uint32_t                rf_lvblen;
 394         uint32_t                rf_lsflags;
 395         uint64_t                rf_unused;
 396 };
 397
 398 struct rcom_lock {
 399         uint32_t                rl_ownpid;
 400         uint32_t                rl_lkid;
 401         uint32_t                rl_remid;
 402         uint32_t                rl_parent_lkid;
 403         uint32_t                rl_parent_remid;
 404         uint32_t                rl_exflags;
 405         uint32_t                rl_flags;
 406         uint32_t                rl_lvbseq;
 407         int                     rl_result;
 408         int8_t                  rl_rqmode;
 409         int8_t                  rl_grmode;
 410         int8_t                  rl_status;
 411         int8_t                  rl_asts;
 412         uint16_t                rl_wait_type;
 413         uint16_t                rl_namelen;
 414         char                    rl_name[DLM_RESNAME_MAXLEN];
 415         char                    rl_lvb[0];
 416 };
 417
 418 struct dlm_ls {
 419         struct list_head        ls_list;        /* list of lockspaces */
 420         dlm_lockspace_t         *ls_local_handle;
 421         uint32_t                ls_global_id;   /* global unique lockspace ID */
 422         uint32_t                ls_exflags;
 423         int                     ls_lvblen;
 424         int                     ls_count;       /* reference count */
 425         unsigned long           ls_flags;       /* LSFL_ */
 426         struct kobject          ls_kobj;
 427
 428         struct dlm_rsbtable     *ls_rsbtbl;
 429         uint32_t                ls_rsbtbl_size;
 430
 431         struct dlm_lkbtable     *ls_lkbtbl;
 432         uint32_t                ls_lkbtbl_size;
 433
 434         struct dlm_dirtable     *ls_dirtbl;
 435         uint32_t                ls_dirtbl_size;
 436
 437         struct mutex            ls_waiters_mutex;
 438         struct list_head        ls_waiters;     /* lkbs needing a reply */
 439
 440         struct list_head        ls_nodes;       /* current nodes in ls */
 441         struct list_head        ls_nodes_gone;  /* dead node list, recovery */
 442         int                     ls_num_nodes;   /* number of nodes in ls */
 443         int                     ls_low_nodeid;
 444         int                     ls_total_weight;
 445         int                     *ls_node_array;
 446
 447         struct dlm_rsb          ls_stub_rsb;    /* for returning errors */
 448         struct dlm_lkb          ls_stub_lkb;    /* for returning errors */
 449         struct dlm_message      ls_stub_ms;     /* for faking a reply */
 450
 451         struct dentry           *ls_debug_rsb_dentry; /* debugfs */
 452         struct dentry           *ls_debug_waiters_dentry; /* debugfs */
 453
 454         wait_queue_head_t       ls_uevent_wait; /* user part of join/leave */
 455         int                     ls_uevent_result;
 456
 457         struct miscdevice       ls_device;
 458
 459         /* recovery related */
 460
 461         struct timer_list       ls_timer;
 462         struct task_struct      *ls_recoverd_task;
 463         struct mutex            ls_recoverd_active;
 464         spinlock_t              ls_recover_lock;
 465         uint32_t                ls_recover_status; /* DLM_RS_ */
 466         uint64_t                ls_recover_seq;
 467         struct dlm_recover      *ls_recover_args;
 468         struct rw_semaphore     ls_in_recovery; /* block local requests */
 469         struct list_head        ls_requestqueue;/* queue remote requests */
 470         struct mutex            ls_requestqueue_mutex;
 471         char                    *ls_recover_buf;
 472         int                     ls_recover_nodeid; /* for debugging */
 473         uint64_t                ls_rcom_seq;
 474         spinlock_t              ls_rcom_spin;
 475         struct list_head        ls_recover_list;
 476         spinlock_t              ls_recover_list_lock;
 477         int                     ls_recover_list_count;
 478         wait_queue_head_t       ls_wait_general;
 479         struct mutex            ls_clear_proc_locks;
 480
 481         struct list_head        ls_root_list;   /* root resources */
 482         struct rw_semaphore     ls_root_sem;    /* protect root_list */
 483
 484         int                     ls_namelen;
 485         char                    ls_name[1];
 486 };
 487
 488 #define LSFL_WORK               0
 489 #define LSFL_RUNNING            1
 490 #define LSFL_RECOVERY_STOP      2
 491 #define LSFL_RCOM_READY         3
 492 #define LSFL_RCOM_WAIT          4
 493 #define LSFL_UEVENT_WAIT        5
 494
 495 /* much of this is just saving user space pointers associated with the
 496    lock that we pass back to the user lib with an ast */
 497
 498 struct dlm_user_args {
 499         struct dlm_user_proc    *proc; /* each process that opens the lockspace
 500                                           device has private data
 501                                           (dlm_user_proc) on the struct file,
 502                                           the process's locks point back to it*/
 503         struct dlm_lksb         lksb;
 504         int                     old_mode;
 505         int                     update_user_lvb;
 506         struct dlm_lksb __user  *user_lksb;
 507         void __user             *castparam;
 508         void __user             *castaddr;
 509         void __user             *bastparam;
 510         void __user             *bastaddr;
 511 };
 512
 513 #define DLM_PROC_FLAGS_CLOSING 1
 514 #define DLM_PROC_FLAGS_COMPAT  2
 515
 516 /* locks list is kept so we can remove all a process's locks when it
 517    exits (or orphan those that are persistent) */
 518
 519 struct dlm_user_proc {
 520         dlm_lockspace_t         *lockspace;
 521         unsigned long           flags; /* DLM_PROC_FLAGS */
 522         struct list_head        asts;
 523         spinlock_t              asts_spin;
 524         struct list_head        locks;
 525         spinlock_t              locks_spin;
 526         wait_queue_head_t       wait;
 527 };
 528
 529 static inline int dlm_locking_stopped(struct dlm_ls *ls)
 530 {
 531         return !test_bit(LSFL_RUNNING, &ls->ls_flags);
 532 }
 533
 534 static inline int dlm_recovery_stopped(struct dlm_ls *ls)
 535 {
 536         return test_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
 537 }
 538
 539 static inline int dlm_no_directory(struct dlm_ls *ls)
 540 {
 541         return (ls->ls_exflags & DLM_LSFL_NODIR) ? 1 : 0;
 542 }
 543
 544 #endif                          /* __DLM_INTERNAL_DOT_H__ */
 545